On Tue, Feb 21, 2012 at 11:49 AM, H. Peter Anvin <hpa@xxxxxxxxx> wrote: > On 02/21/2012 11:40 AM, Andrew Lutomirski wrote: >>> >>> I was having delusions that we could have a task-owned PDT in negative >>> space, but that would require unsharing the third level, too, which is >>> just way too messy. >> >> I'd like to do that, too, and I'd also like to have a per-cpu >> kernel-only page in there, but that's even worse. If we had a >> separate cr3-like register for negative addresses, life would be good >> :) >> > > No, that wouldn't help. The situation is actually quite similar to the > current situation where we have an unshared fourth level, but since the > fourth entries are 512G per entry, we would have to push unsharing of > the kernel address space at least one more level (1G), possibly two > (2M). Painful. > > The main advantage of a separate cr3 would be that we wouldn't need the > unshared top level for the kernel side. Also, as is, if the top level wants to be per-cpu *and* per-task, that's a big explosion of page tables that all need to stay in sync. Oh well. --Andy -- To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
